What attrites when and why: Implic

50 downloads 0 Views 209KB Size Report
Among the complex syntactic characteristics most commonly considered in current research are A-bar movement regularities4 (Lee-Ellis 2011;. O'Grady, et al.
Author’s pre-publication copy To Appear in Oxford Handbook on Language Attrition

Chapter 5: What attrites when and why: Implications of the Bottleneck Hypothesis Roumyana Slabakova University of Southampton and UiT The Arctic University of Norway

Abstract: This chapter explores the predictions of the Bottleneck Hypothesis to language attrition. The hypothesis is based on comparing different degrees of success or difficulty in the acquisition of syntax, semantics and morphology. Its main tenets are that functional morphology presents the biggest challenge to acquisition, while syntax and semantics are relatively easier to acquire because they employ universal operations. The grammars of early and late attriters are examined with a view of checking these expectations. An overview of the literature suggests that early attriters are indeed challenged by inflectional morphology, especially when expressed by large paradigms and when lexical learning of affixes is involved. In contrast, early attriters rarely have issues with basic syntax. Essentially the same picture emerges for late attriters, modulated by linguistic complexity, redundancy of the marker, opaqueness of form–meaning mapping, as well as usage frequency. While the Bottleneck Hypothesis is too large-grained to explain all findings, its predictions appear to be largely borne out. Keywords The Bottleneck Hypothesis, functional morphology, syntax, early attriters, late attriters 5.1 Introduction The Bottleneck Hypothesis (BH, Slabakova, 2008, 2013, 2016) was proposed to identify properties and constructions that are harder or easier for learners to acquire in their second language. It is predicated on the generative linguistics notion attributing a special place to functional morphology in grammar and acquisition. Functional morphemes, both bound

(inflectional) as well as free morphemes, carry grammatical meanings that radically change the overall meaning of a sentence. No sentence meaning is complete without taking the functional morphology contributions into account. The following example illustrates the division of labor between lexical items and functional categories. Suppose we want to combine the lexical items Jane, eat, sandwich into a message with this meaning: “Jane is consuming a sandwich at the moment of speaking.” It is a rare language that would just string the words together with no grammatical morphemes and without taking word order into account. In English, for example, the meanings of COUNT, SINGULAR, NON-FAMILIAR noun are imparted by the indefinite determiner a; the aspectual meaning ONGOING EVENT is reflected in the progressive –ing; the temporal meaning of PRESENT EVENT and agreement with a third person singular subject are captured by the auxiliary is. The final makeup of the sentence Jane is eating a sandwich carries 1

a complete meaning (its truth value) composed of lexical and grammatical meanings that are considerably removed from the raw assembly of lexical verb and nouns. These additional meanings that anchor the event in time and space are contributed by the functional morphemes. The grammatical and semantic features expressed by functional morphology, such as the ones mentioned above as well as DEFINITENESS, SPECIFICITY, etc., are universal in the sense that they have to be expressed in every language, in one way or another. However, languages differ with respect to how these universal meanings are expressed. For example, Mandarin Chinese does not express the past time of a state or an event with a dedicated morpheme, while English does. Korean and Russian are languages without a definite article. The lack of overt exponents does not mean that the respective languages cannot express those meanings, but that they express them with other means, including leaving them for the context to fix. An important feature of functional categories is their multi-functionality. In languages such as English and many others, functional morphemes can encode multiple grammatical meanings. Linguists express this state of affairs by postulating that a number of grammatical features reside (are checked) in the same functional category. In the sentence Gwen often eat-s fish, the agreement morpheme –s signals PRESENT TENSE, HABITUAL ASPECT and agreement with a third person subject. However, two other grammatical meanings are also captured by features in the same functional category Tense: the obligatory subject (*Ø often eats fish) and the lack of I use small caps to set out grammatical meanings with predominantly semantic content, as opposed to purely grammatical meanings such as subject–verb agreement. 1



2

verb movement over the adverb and negation (*Gwen eats often fish, *Gwen eats not fish). Thus, multi-functional Tense in English carries at least five features, three semantic features and two formal syntactic features. In addition to being indispensable in carrying grammatical meanings, the functional morphemes are the main locus of linguistic variation. As per the Minimalist Program assumptions (Chomsky, 1995), the core syntactic operations (e.g., Merge, Agree, etc.) are the same in all natural languages, hence they are also universal just as the grammatical meanings. Parametric differences are no longer attributed to the syntax of two languages differing in constrained ways. If syntactic operations are universal, then variation among languages can only be located in the functional lexicon, the part of the lexicon where the functional morphology resides, and at the two interfaces with meaning and form (or expression). How about semantics? Just like syntax, semantics is postulated to contain only universal operations (e.g., functional application, lambda abstraction, etc.) which allow languages to translate sentence syntactic structure into sentence meaning, or truth value. As mentioned above, grammatical meanings such as TENSE, ASPECT and DEFINITENESS are reflected in the functional lexicon. We can think of each lexical item, including functional and open-class ones, as combining information of three types: semantic, phonological and morpho-syntactic (Jackendoff, 2002). To illustrate with a verb such as drink, when we know the verb, we know that it means “to consume or imbibe a liquid”; we know it is made up of a unique sequence of sounds [drıŋk]; we also know that it is a transitive verb that most of the time takes an object, and that this object may be dropped but is implied (e.g., when used to characterize a person, He drinks does not mean He drinks water). This view of the language architecture (see Slabakova, 2016, chapters 1 and 2) informs views and predictions about language acquisition in general. In addition to learning the equivalents of the native open-class lexical items, acquiring the functional morphology with all its features secures the skeleton of the grammar. Potentially speaking, the three types of information encoded in lexical items can be decoupled in acquisition: learners may not learn them all at the same time. For instance, one can imagine a speaker knowing the basic meaning of drink but not understanding the significance of the implied object (that it can only refer to alcohol). While formal syntactic features are universal, their Universal Grammar-provided values are reflected in the specifications of the functional morphology and need language-specific input



3

to be fixed. Grammatical meanings (TENSE, ASPECT) can also be uncoupled in acquisition, especially if they are complex (PROGRESSIVE and PERFECT ASPECT). For example, the usage of the progressive tense with stative verbs to convey temporary events (He is being honest) may be acquired later than its primary meaning of ongoing event (He is playing). In summary, formal syntactic features and grammatical meanings reflected in a functional category can be dissociated in acquisition, as the features can be learned one by one, or in bundles. While this view of the language architecture is theory-specific, there is experimental evidence to support the dissociation between functional morphology and syntax. White (2003: 182–184), following the idea of morphological separation (Beard, 1987, 1988; Halle & Marantz, 1993), provides convincing evidence for the so-called syntax–before–morphology view. She argues that L2 acquisition data from children and adult learners point to the following dissociation: The morphological expressions of the functional category Tense are being produced with accuracy between around 5% and 45% in obligatory contexts, while the syntactic reflexes of the same functional category (nominative subject, subject remains in VP) are observed with close to a 100% accuracy. It seems that the invisible syntactic knowledge is in place much earlier than the obligatory suppliance of the morphological expressions. Slabakova (2006, 2008) argued that universal semantic knowledge is also in place without much effort on the part of learners. To take a specific example, Slabakova (2015) examined whether learners of L2 Chinese know which sentences are to be interpreted as encoding a past action, if there is no overt morphological expression of Tense. It turns out they do, and even at intermediate levels of proficiency. These and many other studies (see Slabakova, 2008 for review) supply evidence that syntactic and semantic knowledge come before fully accurate realization of the inflectional morphology. In addition, functional morphology is often expressed by complex paradigms (e.g., subject–verb agreement in Slavic and Romance languages), which have to be learned and accessed as separate lexical entries and assembled online both in comprehension and production. Since there is an element of effort-intensive lexical learning in acquiring paradigms, inexperienced learners (children, second language learners) at the beginning of the acquisition process often assume one form to be the representative of the whole morphological paradigm, or a default form. Paradigmatic forms present more difficulty in processing language, as opposed to single form expressions. While there are solid linguistic explanations of why a certain member of a paradigm can become a default, the phenomenon of default forms in principle suggests that not



4

all members of a paradigm are equal, lexically learned at the same time and processed equally easily. When non-default forms are required by the linguistic context, errors are more likely. Finally, some functional morphology has only grammatical functions and not much communicative import. Take for example the agreement between a noun and its modifying adjective. Whether the agreement morpheme is present on the adjective or not, the noun phrase is still comprehensible. Such a situation might apply processing pressure for omissions or supplying default forms. The Bottleneck Hypothesis (BH), whose application to attrition I will explore in this chapter, is based on comparing degrees of success or difficulty in the L2 acquisition of three modules of the grammar: syntax, semantics and morphology. Its main tenets, as specified in (i), (ii), and (iii), are both theoretically motivated and supported with acquisition facts. i. Functional morphology is more challenging for learners than syntax or semantics, which are universal. This claim is founded on the minimalist view that functional morphology encodes variation among languages, and hence is the locus of what has to be acquired. In addition, processing and learning considerations also point to heightened difficulty. ii. The syntactic representations attained by learners may be accurate even though the associated overt morphology may be lacking. iii. The compositional semantic representations employed by learners are accurate. In this chapter, I will apply this language acquisition hypothesis to a different population with the ensuing different outcomes. The purpose is trying to account for language loss, not language gain. Together with recent treatments (see Schmid 2013 for an overview), I will assume that language attrition represents a reanalysis of the grammar over the course of a lifetime in the absence of abundant, consistent input, and in this reanalysis, the language architecture and inherent complexity of grammatical features are an essential factor that affects the process. When a fully acquired language is used under reduced input conditions and under pressure from another language in a bilingual environment, at issue is which areas of grammar remain resilient and likely to be spared, and which ones are vulnerable to attrition? The BH can provide a useful lens to these issues.



5

5.2 A fundamental division based on age: the maturational factor Recent research on language attrition in the lexicon as well as in grammar has led to the conclusion that we can no longer treat language attrition as a unitary phenomenon. In addition to the high amount of individual variation due to personal, motivational and experiential factors, it appears that the age of onset of bilingualism is of utmost importance. With all the necessary caveats needed when we average over a multitude of people’s linguistic experiences, we may be describing two distinct processes. One would be the heritage speaker experience: child learners growing up in homes where the native language is spoken, but in countries with another majority language. Very commonly, the heritage language is the individual’s first language chronologically, a mother tongue learned in a naturalistic setting in a family environment. The majority language comes in second. However, being used at school, with peers and in the wider society, the majority language often becomes dominant and has a dramatic effect over the native language, setting off language attrition (Benmamoun, Montrul & Polinsky, 2010; Montrul, 2008). Broadly defined, the exposure to the dominant societal language comes either at school age or at the age of immigration, but crucially before the age of puberty (around 11) (see chapters $$, $$ in this handbook, as well as Kupisch & Rothman (2016), for more discussion). In our context, these speakers could be called “early attriters.” Summarizing recent research, Benmamoun et al. (2010) argue that loss of functional structure is evident in this population. Another population exhibiting language attrition is adult immigrants themselves. Very often, they are the parents of the heritage language speakers described above. Exposure to the second language is sustained and intense, and may be accompanied with a sharp decrease in exposure to the native language. In our context, the defining characteristic of this group is that they have left the country where their mother tongue is spoken after the age of puberty, and sometimes much later. This ensures that their native language grammar was completely mature and stable before the onset of (potential) attrition. For the purposes of this chapter, I will call these “late attriters.” The current evidence suggests that changes in the core grammar, or in the functional category structure, is not much in evidence in this population. So, while we are pondering the issue of why it is the case that certain areas of the grammar are so resilient whereas other areas are malleable and subject to change throughout the life of a speaker, we



6

should also keep in mind that the age of onset of attrition can change the answer dramatically (Montrul, Bhatt & Girju, 2015; Schmid, 2014). 5.2.1 The grammatical competence of early attriters Heritage speaker competence and performance are notoriously diverse and fluid, so generalizations are to be attempted with caution. Nevertheless, generalizations are indispensable in checking theoretical predictions. Benmamoun et al. (2010), a white paper digesting decades of research on heritage language speakers, summarizes and interprets data from several areas where heritage speakers appear to diverge from monolingual native speakers: phonology, lexical knowledge, morphology, syntax, case marking, and code-switching. In order to check the predictions of the BH against these pooled findings, we shall examine the areas of functional morphology and syntax in some more detail. The following citation comes from an executive summary of the paper on the National Heritage Language Resource Center website. “Morphology, particularly inflectional morphology, is an area in which heritage speakers often have the most trouble [emphasis mine]. Preliminary research suggests some generalizations: They appear to struggle more with nominal morphology than verbal morphology (although there are differences as well within verbal morphology); agreement, aspect, and mood are much weaker areas in heritage speakers than is tense.” http://nhlrc.ucla.edu/nhlrc/resources/article/122865 Agreement is the functional feature that undergoes the highest erosion. A number of studies have demonstrated significant loss of subject–verb agreement in heritage speakers (as compared to monolinguals): Albirini et al. (2011), Anderson (2001), Håkansson (1995), Polinsky (2006), Rothman (2007). For example, Albirini et al. (2011) recorded Egyptian and Palestinian heritage speakers with English as the dominant language and found up to 6.5% of verb agreement errors and up to 21% nominal agreement errors. Speakers experienced particular difficulty in creating agreement chains between the verb and plural or feminine subjects, as well as where more arcane rules of grammar dictate, for example, singular agreement with postverbal



7

conjoined subjects (p. 285). Another example comes from Hungarian, where the possessed noun 2

morphologically agrees with the possessor in person and number. Bolonyai (2007) reports an error rate of 66% for this property among Hungarian speaking children in the US between the ages of 7 and 9. She explicitly argues for the selective nature of morphological variability. Erosion of agreement is particularly striking given that other verbal categories seem to be more resilient, but is predicted by the BH. It is particularly striking that high percentages of agreement errors occur in null subject languages (Arabic in this case), where the agreement affix identifies the missing subject. The lack of semantic significance explanation presents a good account of the Hungarian agreement attrition. Since the possessed and the possessor are both present in the noun phrase, dropping the agreement morpheme does not incur communication loss. In addition to agreement, other functional categories morphologically expressed on the verb also create difficulties. For example, aspect vulnerability has been studied widely in heritage Russian, because aspect is such a prominent feature of the language and because it is intricately linked to lexical knowledge in a complex way (Slabakova, 2005). Russian grammatical aspect is a functional category encoding purely grammatical features but their expression is an amalgamation with derivational morphology. Perfective-marking prefixes comprise some 18 to 20 derivational morphemes, and prefix–verb combinations have to be learned individually. Among many others, Laleko (2011) described aspectual errors that heritage Russian speakers made in free oral production, and while they can be classified as aspectual errors, they also constitute wrong lexical choices. In order to understand the process, the reader must keep in mind that lexical aspect is a semantic property of verbal predicates including the telicity distinction. Unlike lexical aspect, grammatical aspect is expressed through inflectional morphology on the verb (Smith, 1991).

3

It is also interesting to note that there were significant differences among the heritage Arabic speakers. Egyptian Arabic speakers showed a greater loss of agreement than Palestinian Arabic speakers. This divergence may have both structural and sociolinguistic explanations. 2

To eat a sandwich is a telic VP because it has a potential endpoint, while She was eating a sandwich is a clause encoding imperfective viewpoint aspect, since the speaker focuses attention on the process, not on the endpoint of the event. English sentences mark both lexical aspect (telicity) and viewpoint aspect. 3



8

Polinsky (1997, 2008a, 2009) proposed the Lexical Aspect Hypothesis (see also Pereltsvaig, 2005), according to which verbal aspectual morphology (i.e., perfective prefixes) in heritage Russian encodes lexical, rather than grammatical aspect distinctions: “most verbs become either lexicalized perfectives or lexicalized imperfectives” (1997, p. 384). This functional lexicon reanalysis predicts that telic verbs will mostly appear in perfective forms while atelic verbs will surface as imperfectives. These predictions were supported by a vocabulary translation experiment (Polinsky, 1997) and by a forced production task (Laleko, 2011). The BH would explain this lexicalization process by the close link between the lexicon and functional morphology, and the need for these particular functional morphemes to be learned one by one with their appropriate verbs (Slabakova, 2005). Early attriters simplify their lexicon by fusing the prefix and the verbal root. Even if attriters know that a prefixed verb signals a complete event and treat the prefix as separable from the root, they still might produce the wrong prefix for any verb. Staying among verbal inflectional distinctions, it is to be noted that some are much more vulnerable than others. As discussed above, the challenge of Russian aspect is based on its lexical rather than grammatical difficulty. Frequency of usage may be another factor explaining differential difficulty. Let us compare tense with subjunctive mood in Romance, which differ in frequency of occurrence. While both grammatical features are expressed with complex paradigms, tense occurs in every sentence whereas subjunctive mood is rarer and needs a specific context (Iverson, Kempchinsky & Rothman, 2008). Not surprisingly, while mood distinctions are subject to significant attrition (e.g., subjunctive in heritage Spanish), tense distinctions are quite robust, even in a weak heritage language (Montrul, 2009). Next, we shall look at nominal morphology, which does not have much semantic effect on the interpretation of the sentence: gender, case and differential object marking. In some languages, such as Slavic, Germanic and Romance, all nouns are categorized in lexical classes of grammatical gender. Note that this categorization does not carry any informational function, because most nouns refer to inanimates, so gender marking is not linked to natural sex but is purely grammatical. Gender represents another vulnerable feature illustrating the lexical burden of functional morphology, both as noun categorization and as gender agreement within the noun phrase. Montrul, Foote and Perpiñán (2008a) found age-related effects in the marking of heritage Spanish gender agreement, while Polinsky (2008b) discovered that heritage Russian speakers



9

had restructured the formal cues used to assign Russian nouns to gender classes. In the latter study, this restructuring was related to general proficiency in Russian. Case marking also displays attrition and loss, especially in constructions that allow optionality, such as the genitive of negation in Russian (Leisiö, 2006). Differential object marking in Spanish, a grammatical phenomenon with complex semantic distinctions (see examples in later sections), is frequently deficient among heritage speakers (Montrul & Bowles, 2009). In sum, early attriters are challenged by inflectional morphology when it is expressed by large paradigms involving affix learning and online assembly (agreement), when marking a grammatical category implicates a lexically fixed affix–verb combination (aspect) or individual exponence (gender) and when the distinction is rarely used (mood). In contrast to difficulties with functional morphology, early attriters rarely have issues with basic syntax, such as word order in simple clauses or verb movement. There is not much work on low-level syntactic properties because researchers focus their attention on much more complex syntax, but at least some evidence on accurate verb raising in heritage Spanish is available (Bruhn de Garavito, 2002). Among the complex syntactic characteristics most commonly considered in current research are A-bar movement regularities (Lee-Ellis 2011; 4

O’Grady, et al. 2001; O’Grady, et al., 2011; Polinsky, 2011), wh-movement (Montrul, Foote & Perpiñán, 2008b) and binding (Keating et al., 2011; Kim et al., 2009). An interesting case in point is presented by Montrul et al. (2008b), which studied a range of interrogative sentences and extractions out of main and embedded clauses. The researchers tested early bilinguals at three proficiency levels and compared them to similar proficiency adult-onset L2 learners of Spanish. Both the L2 learners and the heritage speakers demonstrated solid knowledge of constraints on wh-movement, even in constructions where transfer from English would lead the speakers astray. The only area where they differed from native speakers was subject and object extraction from embedded clauses. These findings were interpreted as pointing to correct underlying representations in all L2 learners and heritage speakers but difficulties in processing such long sentences. The general conclusion of this and comparable studies is that basic syntactic operations are not vulnerable to attrition even in early bilinguals, while their reduced practice of the heritage language may lead to superficial processing deficiencies. A-bar movement involves extraction of a phrase from its position of merging and case marking to another position in the sentence, usually to the left periphery of the sentence. 4



10

How about knowledge of semantics and appropriateness in discourse context? The BH predicts that comprehension resulting from semantic composition of already established grammatical features and constructions is not going to be vulnerable. Studies on such properties are rare, but they do exist. Leal Méndez, Rothman and Slabakova (2014) tested whether heritage speakers at two proficiency levels and adult L2-ers recognize the meaning and appropriate context of clitic left dislocations (CLLD) and focus fronting in Spanish. The English equivalent to CLLD, topicalization, does not double the dislocated object with a clitic. CLLD is a very frequent feature of Spanish. The specific semantic constraint tested was the semantic freedom of the antecedent–dislocate connection: the dislocated clitic-doubled constituent can be identical, a subset or even a part of the discourse antecedent. This semantic freedom is universal for all topicalizations, hence the BH would predict that early attriters would not have any problem with comprehending it. The example below illustrates the part–whole relationship between the dislocated constituent (the table legs) and the antecedent (the table). (1)

Context: What shall we do with the table? It is too big! Mira, las patas, las

doblas

look the legs Cl.ACC fold.PRES.2sg

Part/whole

así… this-way

‘Look, you can fold the legs like this…’ While attriters might make errors with the grammar of CLLD by not providing the clitic doubling, they should not be less accurate on sentences such as (1) compared to sentences in which the dislocated element and its antecedent are identical. A felicity judgment task ascertained that the learners and heritage speakers were no different from the monolingual natives in their comprehension of this semantic freedom. The Interface Hypothesis (IH, Sorace, 2011) was initially proposed to address language attrition and later expanded to other bilingual conditions (see chapter 4 of this handbook). The predictions of the IH and the BH differ: Whereas the former hypothesis sees most challenge at the interface of grammar and external modules such as context, the latter hypothesis places the main burden of difficulty on the functional morphology. While there is substantial evidence for the IH coming predominantly from null and overt subject usage, a range of experiments on other interface constructions has not encountered heritage speaker deficiencies, challenging its



11

predictions. For example, the same experiment cited above, Leal Méndez et al. (2014), in addition to checking a semantic constraint, also established that the heritage speakers had no problem with choosing the context-appropriate construction with the clitic doubling. Furthermore Leal, Rothman and Slabakova (2014) looked at a very rare construction, clitic right dislocation, which is only acceptable in highly circumscribed contexts such as the affective context illustrated below. Note that the object is right-dislocated in (2a), but is doubled by a preverbal clitic: (2) Context: Mariana and Omar are siblings and they inherited several pieces of furniture after their father died. Because Mariana has a larger house, she kept the majority of them. The problem is that many pieces of furniture were damaged, and Mariana initially didn’t know what to do with them. One day, Omar goes to visit her; he thinks he recognizes a chair in the living room. O:

Veo que por fin decidiste arreglar la silla de papá. ‘I see that you finally decided to fix dad’s chair.’

(a)

M:

No. La regalé, la silla odiosa esa. Lo que arreglé fue el sofá que tanto me

gustaba. no CL I-gave the chair ugly that. CL that fixed was the sofa that so me pleased (b)

M:

No. # Regalé, la silla odiosa esa. Lo que arreglé fue el sofá que tanto me

gustaba. no I-gave the chair ugly that. CL that fixed was the sofa that so me pleased ‘No, I gave that hideous chair away. What I did repair was the sofa that I liked so much.’ The experiment used a judgment task where speakers had to rate the felicity of (2a) or (2b) in the given context. Results established that Spanish-dominant and heritage bilinguals were equally selective with the contexts in which they allowed clitic right dislocation. No group differences were in evidence. The researchers argued that these findings challenge the Interface Hypothesis, at least in off-line tasks where there are no time constraints. To bring all these disparate threads together, the comparison between early attriters’ knowledge of functional morphology, basic syntactic operations and semantic composition



12

constitutes support for the BH, in the sense that functional morphology is much more vulnerable to attrition than the latter two areas of the grammar. However, in summarizing apparent knowledge of different properties, it also emerged that grouping linguistic properties and their acquisition on the basis of gross modules may be over-simplistic. Some functional morphology such as agreement seems to be more vulnerable than other functional morphology such as tense marking, with various features in between these two extremes. In other words, the functional morphology is not monolithic but can be selectively attrited. A tentative explanation based on the grammatical status, complexity of the grammatical features and the evidence for their expression runs like this: If a grammatical feature does not affect meaning-making and comprehension, but captures purely syntactic effects (such as subject–verb agreement and differential object marking), then this feature is vulnerable. Seemingly redundant morphology is not retained in heritage grammars but simplified, possibly leading to a restructuring of the overall grammar (see also Seliger 1996). If a piece of functional morphology has a semantic or discourse-pragmatic effect and the evidence for it is rarely seen in the language, then that feature may also be vulnerable (mood, differential object marking). If functional morphology is conflated with derivational morphology as in Russian aspect, it is likely to be restructured. One-to-one form– meaning mappings that are clearly and abundantly evidenced in the linguistic input (such as tense marking) endure the best. This explanation is compatible with the spirit of the Interpretability Hypothesis (Tsimpli & Dimitrakopoulou, 2007) because it elevates meaningrelated features as more resilient than form-related ones.

5

5.2.2 The grammatical competence of late attriters Let us now turn our attention to later attriters, the immigrants who moved to the country of their L2 well after puberty. Studies documenting attrition of functional morphology among later 5 The Interpretability Hypothesis (Tsimpli &Dimitrakopoulou 2007) has been proposed to explain the inability of second language learners to reach nativelike levels of competence. The hypothesis builds on the Minimalist Program interpretable–uninterpretable feature distinction and postulates that interpretable features are acquirable in the L2 while purely syntactic uninterpretable features are not, if they are not already established in the native language and thus transferable. Note also the explanation provided above is in contrast to that proposed by Tsimpli, Sorace, Heycock and Filiaci (2004), which gives rise to the Interface Hypothesis.

13

attriters are rare, but do exist. For example, Ribbert and Kuiken (2010) documented attrition of the infinitival complementizer among Germans who moved to the Netherlands as university students. There is both overlap and difference in the use of Dutch and German infinitival complementizer um/om: Dutch om is a superset of the contexts where German um is used. Furthermore, Dutch has a high number of contexts where om can be used optionally. The German speakers were asked to rate the grammaticality of pairs of sentences with and without a complementizer. Their preferences differed significantly from other German speakers who were not in contact with the Dutch language. Still staying within functional morphology and its interpretation, selective attrition is attested in late attriters, just like in early attriters. Gürel (2004) provides a very interesting example. She looked at a rare morpho-syntactic phenomenon with semantic effects: the binding of overt, null and reflexive pronouns in Turkish embedded clauses as exemplified in (3a) below. The overt third person pronoun o, when it is embedded clause subject, cannot refer back to the main clause subject, while the null pronoun (pro) can. The English overt pronoun he in (3b) has no such restriction. In the examples, the sub-index “i” refers to the main subject while “j” means that the pronoun has some other discourse-provided antecedent. (3a)

Burak [o-nun / kendi-si-nin / pro zeki i

*i/j

i/j

i/j

ol-duğ-u]-nu

düşün-üyor

Burak s/he-GEN self-3SG-GEN pro intelligent be-NOM-3SGPOSS-ACC think-PRG ‘Burak thinks that [he / self / pro is intelligent]’ i

(3b)

*i/j

i/j

i/j

Burak thinks that he is intelligent. i

i/j

In two interpretation tasks where speakers had to choose between the pronoun antecedent being the main clause subject and a third person not mentioned in the sentence, native speakers overwhelmingly chose the third person, the only available interpretation. However, Turkish attriters living in the USA and Canada chose the wrong antecedent between 21% and 30% (on the different conditions). They still preferred the third person as antecedent, choosing it above 70%; however, their knowledge in this respect was evidently shaky. In contrast to the overt pronoun o, the attriters maintained their knowledge of the nominative reflexive kendisi and the null pronoun. It is also interesting to note that this selective attrition is paralleled exactly by selective acquisition among near-native L2 learners of Turkish (Gürel, 2006). It appears that the



14

null pronoun and kendisi are acquired completely and resist attrition, while the antecedent choices of o in embedded subject position create problems for both near-native speakers and late attriters. Binding is among the most complex properties of language, relying on syntactic features captured by functional morphemes with semantic import, and frequently discoursepragmatic import as well. The prohibition of main clause antecedents for the overt pronoun o, but not for the other two pronominal elements, is eroded under the influence of English where such prohibition does not exist. In one sense, this restructuring is a simplification of the grammar. In checking the predictions of the BH, it is informative to look at Schmid (2014), a study comparing a number of late attriters with very advanced second language speakers, matched in proficiency and looking at a number of linguistic properties. The attriters were German native speakers living in Canada, while the advanced L2 learners and controls were living in Germany at the time of testing. The data comes from two spontaneous speech samples per subject, a film retell task and a semi-structured autobiographical interview, in which errors were calculated per 1,000 words. The range of functional morphemes tested included case, gender, plural, articles, pronouns, tense morphology (the weak/strong distinction), auxiliaries, and subject–verb agreement; while syntactic properties included verb-second and discontinuous word order, as well as word order in subordinate clauses. Schmid reported the error counts for each speech sample separately (her Table 5 on p. 401). The late attriters were non-distinguishable from the monolingual native speakers for all of the properties tested, functional morphology and syntax alike. The BH cannot offer an explanation for their linguistic performance just because there is nothing to explain: these German immigrants to Canada did not demonstrate any attrition. In contrast to the findings in Schmid (2014), Montrul et al. (2015) did attest attrition of functional morphology. This a large-scale study of differential object marking (DOM) examined in heritage speakers and their parent generation, thus early and late attriters, of Spanish, Hindi/Urdu and Romanian. DOM is among the most complicated functional morphology, where object marking depends mainly on animacy and specificity, but aspect, topicality, agentivity and affectedness may also influence it. The example in (4) comes from Hindi. (4)



Mira-ne ramesh-ko

dekhaa.

Mira-erg Ramesh-DOM

saw

15

‘Mira saw Ramesh.’ (*Mira-ne ramesh dekhaa.) The three languages under investigation work largely (although not entirely) in parallel with respect to DOM’s semantic restrictions, but while the Spanish marker is a (not audible after a vowel), Hindi and Romanian markers are syllabic (ko and pe, respectively), so better audible. The findings are complicated. The heritage speakers accepted ungrammatical DOM omission with animate and specific objects to higher degrees than they accepted omission of indirect object or locative marking. Thus, they were simplifying their grammar in the direction of redundancy elimination (compared to a locative, DOM is largely redundant, as context certainly distinguishes specific from non-specific objects). Thirteen (out of 21, 62%) late attriters of Mexican Spanish had lost DOM in the US, while Hindi and Romanian speakers were not attrited for this property at all. While these results are difficult to explain straightforwardly, the linguistic complexity of the phenomenon is the prime suspect. In addition to the functional morphology being largely redundant, the markers are multi-functional in the sense that they express other grammatical meanings as well, producing an “opaque form–meaning mapping” (O’Grady et al., 2011; Slabakova, 2009). The phonological salience mentioned above cannot be the sole factor explaining the group differences. The authors make a convincing argument that some languageinternal structural factors such as clitic-doubling in Romanian and lack of definite articles in Hindi may be a cause of differences. Ultimately, DOM markers have to be considered as functioning in tandem with the rest of the grammar and being affected by the holistic linguistic experience. The authors conclude that “the structural properties of the DOM markers together with the syntax of definiteness and specificity in each language (more than external factors) seem to account for the degree of DOM erosion in each language” (Montrul et al., 2015: 604). Turning to knowledge of syntax and of interpretation now, the findings of an influential study, Tsimpli et al. (2004) are worth reviewing here. The study investigated production and comprehension of overt and null preverbal and postverbal subjects among late attriters of Italian and Greek living in an English-speaking environment. A robust finding was that the syntax of subjects and interpretation of null subjects were not attrited, while overt subject interpretation choices were more vulnerable because they depended on consideration of discourse-pragmatics



16

as well as syntax. These and other findings were conceptualized as the Interface Hypothesis (see more on this hypothesis in chapter 4). Another recent study, Flores (2012), compared knowledge of syntax and discourse in four age groups of German-Portuguese bilinguals living in Portugal, for whom German was the minority language. Flores’s subject pool allowed her to make very precise distinctions based on 6

age of loss of contact with the attriting language. She examined a narrow syntax property, verb placement in main and embedded clauses, and a syntax–discourse interface property, object expression under topichood. The speakers who had lost contact with German before the age of 10 exhibited around 45% errors on both properties. Those who had moved to Portugal between the ages of 11 and 15 were markedly different: while their errors on topic object drop were in the 35% range, their word order errors were negligible. The third group of adult returnees (at ages15–21) and the group of children who were still in contact with German made no errors on either property. The author concluded that narrow syntax was more likely to suffer from attrition if contact with German was lost before age 10-11, while object expression continues to be vulnerable to reduced input until the age of 15. Late attriters did not exhibit any attrition of these properties. Let us take stock of the main findings for late attriters. While attrition of grammatical features is much rarer among this population (Schmid, 2014), some functional morphology does suffer attrition (Ribert & Kuiken, 2010; Montrul et al., 2015). Purely syntactic properties such as verb placement (Flores, 2012) and the syntax of null and overt subjects (Tsimpli et al., 2004) are much more robust and do not appear to undergo attrition. Finally, interpretive knowledge seems to be vulnerable only if it is calculated at the syntax-discourse interface (Tsimpli et al., 2004). At the same time, Flores’s (2012) adult attriter group did not demonstrate such vulnerability. The Bottleneck Hypothesis is largely borne out by such findings. If anything is vulnerable in the grammars of late attriters, it is the functional morphology and not pure syntax. The additional factors we saw at work in early attriters, namely linguistic complexity of the property under investigation, redundancy of the functional morphology marker, opaqueness of form–meaning mapping, as well as usage frequency hence activation, go a long way to explaining most of the findings we discussed in this section. At the same time, the hypothesis, as 6 Flores’s child control group was the exception, because these subjects were still living in German-speaking countries.

17

it is formulated now, may be too large-grained to provide a sufficient explanation of particular selective attrition, such as the one discovered by Gürel (2004), where a restriction of one pronominal form is relaxed, or by Montrul et al. (2015), where one group of late attriters is affected but not others. 5.3

Conclusion

Examining findings from early and late attrited speakers invites fundamental questions on what the grammar of these speakers look like. In other words, what are the linguistic representations that allow not altogether erroneous, but “shaky” or “rusty” performance on some properties? The BH is useful in identifying the essential grammatical complexity of the functional morphology, especially with many-to-many features-to-form mappings, as the main reason. In comparison, the core syntactic operations and the compositional principles of putting the meaning of a message together are resilient in the face of sub-optimal input, just because they are universal and are used in the dominant L2 as well as in the affected L1. While the BH addresses the issue of why certain properties are differentially vulnerable in attrited grammars, a related important question is how this state of grammatical knowledge operates in individual speakers’ mind/brain. Reminiscent of Yang’s (2002) variationist learning approach, a recent generative proposal, Amaral and Roeper (2014), specifically endorses and explains the mechanism of how all the grammars of a multilingual can arise and function in tandem. “[A]ny language contains properties of several recognizable language types, i.e. the grammar of a language L1 can have elements that form sub-grammars compatible with L2, L3, Ln” (Amaral & Roeper 2014, p. 4). According to this view, L2/Ln acquisition is a natural proliferation of sub-grammars, a process of grammar expansion that is certain to affect the native language right back. Multilingual linguistic competence is an amalgamation of sub-grammars coming from the previously acquired languages, equipped with some sort of differentiation mechanism, such as a tagging device for the rules and lexicon of grammar A or grammar B. If one rule, in the attrition situation the L2 rule, is used much more intensively, its activation level is likely higher in the sense that it will be invoked much faster and with less effort. The L1 rule is not truly gone and it is still tagged as a L1 rule but summoning it may be more effortful and costly. This is essentially a processing view of grammar usage (see also Sharwood Smith & Truscott, 2014), which explains how grammatical rules, while they are in place, may appear “rusty” and “shaky” in performance.



18

References: Albirini, A., Benmamoun, E. & Saadah, E. 2011. Grammatical features of Egyptian and Palestinian Arabic heritage speakers’ oral production. Studies in Second Language Acquisition, 33, 273–303. Amaral, L., & T. Roeper (2014). Multiple Grammars and Second Language Representation Second Language Research, 30, 1, 3–36. Anderson, R. T. (2001). Lexical morphology and verb use in child first language loss: A preliminary case study investigation. International Journal of Bilingualism, 5, 377–401. Beard, R. (1987). Morpheme Order in a Lexeme/Morpheme-Based Morphology, Lingua 72, pp. 1–44. Beard, R. (1988). On the separation of derivation from morphology toward a lexeme/morphemebased morphology, Quaderni di Semantica, 9(1), 3–59. Benmamoun, A., Montrul, S. & Polinsky, M. (2010). Prolegomena to heritage linguistics. Unpublished white paper. University of Illinois at Urbana-Champaign and Harvard University, available from http://scholar.harvard.edu/mpolinsky/publications/whitepaper-prolegomena-heritage-linguistics Bolonyai, A. (2007). (In)vulnerable agreement in incomplete bilingual L1 learners. International Journal of Bilingualism, 11, 3–21. Bruhn de Garavito, J. (2002). Verb raising in Spanish, a comparison of early and late bilinguals. In Proceedings of the 26th annual Boston University Conference on Language Development. Edited by Barbora Skarabela, Sarah Fish, and Anna H.-J. Do, 84–94. Somerville, MA: Cascadilla. Chomsky, N. (1995). The Minimalist Program. Cambridge, MA: MIT Press. Flores, C. 2012. Differential effects of language attrition in the domains of verb placement and object expression. Bilingualism: Language and Cognition, 15, 550–567. Gürel, A. (2004). Selectivity in L2-induced L1 attrition: A psycholinguistic account. Journal of Neurolinguistics 17(1): 53-78. Gürel, A. (2006). L2 acquisition of pragmatic and syntactic constraints in the use of overt and null subject pronouns. In Slabakova, R., Montrul, S and Prevost, P. (Eds.), Inquiries in Linguistic Development (Pp. 259-282). Amsterdam: John Benjamins.



19

Håkansson, G. (1995). Syntax and morphology in language attrition: A study of five bilingual expatriate Swedes. International Journal of Applied Linguistics, 5, 153–169. Halle, M. & Marantz, A. (1993). Distributed Morphology and the Pieces of Inflection. In The View from Building 20, ed. by Kenneth Hale and S. Jay Keyser, pp. 111– 176. Cambridge, MA: MIT Press. Iverson, M., Kempchinsky, P. & Rothman, J. (2008). Interface vulnerability and knowledge of the subjunctive/indicative distinction with negated epistemic predicates in L2 Spanish. EUROSLA Yearbook 8, 135–163. Jackendoff, R. (2002). Foundations of language. Oxford: Oxford University Press. Jakobson, R. (1941). Kindersprache, Aphasie und allgemeine Lautgesetze. Uppsala: Almqvist & Wiksell. Keating, G. D., VanPatten, B. & Jegerski, J. (2011). Who was walking on the beach? Anaphora resolution in Spanish heritage speakers and adult second language learners. Studies in Second Language Acquisition, 33(2), 193–221. Keijzer, M. (2007). Last in first out? An investigation of the regression hypothesis in Dutch emigrants in Anglophone Canada. Vrije Universiteit Amsterdam, PhD dissertation. Kim, J.-H., Montrul, S., & Yoon, J. (2009). Binding interpretation of anaphors in Korean heritage speakers. Language Acquisition, 16(1), 3–35. Laleko, O. (2011) Restructuring of verbal aspect in Heritage Russian: Beyond lexicalization International Journal of Language Studies, 5(3), 13–26. Leal Méndez, T., Rothman, J & Slabakova, R. (2014). Discourse-sensitive clitic-doubled dislocations in heritage Spanish. Lingua, 155, 85–97. Leal, T., Rothman, J. & Slabakova, R. (2014). A Rare Structure at the Syntax-Discourse Interface: Heritage and Spanish-Dominant Native Speakers Weigh In, Language Acquisition, 21(4), 411–429. Lee-Ellis, S. (2011). The elicited production of Korean relative clauses by heritage speakers. Studies in Second Language Acquisition, 33, 57–89. Leisiö, L. (2006). Genitive subjects and objects in the speech of Finland Russians. Journal of Slavic Linguistics 14:289–316.



20

Kupisch, T. & Rothman, J. (2016, articles in press). Terminology matters! Why difference is not incompleteness and how early child bilinguals are heritage speakers International Journal of Bilingualism https://doi.org/10.1177/1367006916654355 Montrul, S. (2008). Incomplete Acquisition in Bilingualism. Re-examining the Age Factor. Amsterdam: John Benjamins. Montrul, S. (2009). Knowledge of tense-aspect and mood in Spanish heritage speakers. International Journal of Bilingualism, 13, 239–269. Montrul, S. & Bowles, M. (2009). Back to basics: Differential object marking under incomplete acquisition in Spanish heritage speakers. Bilingualism: Language and Cognition, 12, 363–383. Montrul, S., Foote, R. & Perpiñán, S. (2008a). Gender agreement in adult second language learners and Spanish heritage speakers: The effects of age and context of acquisition. Language Learning, 58, 503–554. Montrul, S., Foote, R. & Perpiñán, S. (2008b). Knowledge of wh-movement in Spanish L2 learners and heritage Speakers. In Selected Proceedings of the 10th Hispanic Linguistics Symposium, edited by Joyce Bruhn de Garavito and Elena Valenzuela (pp. 93–106). Somerville, MA: Cascadilla Press. Montrul, S., Bhatt, R. & Girju, R. (2015). Differential object marking in Spanish, Hindi and Romanian as heritage languages. Language, 91(3), 564–610. O’Grady, W, Lee, M. & Choo, M. (2001). The acquisition of relative clauses by heritage and non-heritage learners of Korean as a second language: A comparative study. Journal of Korean Language Education, 12, 283–294. O’Grady, W., Kwak, H-Y., Lee, M. & Lee, O-S. (2011). An emergentist perspective on partial language acquisition. Studies in Second Language Acquisition 33, 223–46. Pereltsvaig, A. (2005). Aspect lost, aspect regained: Restructuring of aspectual marking in American Russian. In P. Kempchinsky and R. Slabakova (Eds.), Aspectual inquiries (pp. 369–396). Dordrecht: Kluwer. Polinsky, M. (1997). American Russian: Language loss meets language acquisition. In Brown, W., Dornisch, E., Kondrashova, N., & Zec, D. (Eds.), Formal Approaches to Slavic Linguistics 4 (370-406). Ann Arbor: Michigan Slavic Publishers.



21

Polinsky, M. (2006). Incomplete acquisition: American Russian. Journal of Slavic Linguistics, 14, 191–262. Polinsky, M. (2008a). Russian gender under incomplete acquisition: Heritage speakers’ knowledge of noun categorization. Heritage Language Journal 6.1. 40–71. Online: http://www.heritagelanguages.org. Polinsky, M. (2008b). Without aspect. In Greville G. Corbett and Michael Noonan (Eds.), Case and grammatical relations: Studies in honor of Bernard Comrie (pp. 263–282). Amsterdam: John Benjamins. Polinsky, M. (2009). Without aspect. In Corbett, G., & Noonan, M. (Eds.), Case and Grammatical Relations (263–282). Amsterdam: John Benjamins. Polinsky, M. (2011). Reanalysis in adult heritage language: A case for attrition. Studies in Second Language Acquisition, 33(2), 1-45. Ribbert, A. & Kuiken, F. (2010). L2-induced changes in the L1 of Germans living in the Netherlands. Bilingualism: Language and Cognition, 13(1), 41–48. Rothman, J. (2007). Heritage speaker competence differences, language change, and input type: Inflected infinitives in heritage Brazilian Portuguese. International Journal of Bilingualism, 11, 359–389. Schmid, M. (2013). First language attrition. Linguistic Approaches to Bilingualism, 3(1), 94– 115. Schmid, M. (2014). The Debate on Maturational Constraints in Bilingual Development: A Perspective from First-Language Attrition, Language Acquisition, 21(4), 386-410. Seliger, H. (1996). Primary language attrition in the context of bilingualism. In T. J. Bhatia, & W. C. Ritchie (Eds.), Handbook of second language acquisition (pp. 605–626). New York: Academic Press. Sharwood Smith, M., & Truscott, J. (2014). The multilingual mind: A modular processing perspective. Cambridge: Cambridge University Press. Slabakova, R. (2005). What is So Difficult about Telicity Marking in L2 Russian? Bilingualism: Language and Cognition, 8, 63-77. Slabakova, R. (2008). Meaning in the Second Language, Berlin: Mouton de Gruyter. Slabakova, R. (2009). Features or Parameters: which one makes SLA easier, and more interesting to study? Second Language Research, 25, 2, 313-324.



22

Slabakova, R. (2013). What is easy and what is hard in second language acquisition: A generative perspective. In María del Pilar García Mayo, M. Junkal Gutierrez-Mangado & María Martínez Adrián (Eds.), Contemporary Approaches to Second Language Acquisition (pp. 5–28). Amsterdam: John Benjamins. Slabakova, R. (2015). Acquiring temporal meanings without tense morphology: The case of L2 Mandarin Chinese. Modern Language Journal, 99, 2, 283–307. Slabakova, R. (2016). Second Language Acquisition. Oxford: Oxford University Press. Smith, C. (1991). The parameter of aspect. Dordrecht: Kluwer. Sorace, A. (2011). Pinning down the concept of “interface” in bilingualism. Linguistic Approaches to Bilingualism, 1, 1–33. Tsimpli, I. & Dimitrakopoulou, M. (2007). The interpretability hypothesis: Evidence from whinterrogatives in second language acquisition. Second Language Research, 23(2), 215– 242. Tsimpli, I., Sorace, A.,Heycock, C. & Filiaci, F. (2004). First language attrition and syntactic subjects: A study of Greek and Italian near-native speakers of English. International Journal of Bilingualism, 8(3), 257–277. White, L. (2003). Second Language Acquisition and Universal Grammar. Cambridge: CUP Yang, C. (2002). Knowledge and learning in natural language. New York: Oxford University Press.



23